[ofw] RE: Bad merge in checking 1440

[Fab Tillier]

> From: Sean Hefty [mailto:sean.hefty at intel.com]
> Sent: Wednesday, July 23, 2008 3:08 PM
>
> To help, I included this comment when I sent out the patch for review:
>
> As far as I could tell, there is no code that ever results in passing
> the work request structure between userspace and the kernel.

UAL_POST_SEND and UAL_POST_RECV IOCTLs, issued from ual_post_send and ual_post_recv functions in core\al\user\ual_qp.c

The ib_wc_t structure is passed in the UAL_POLL_CQ IOCTL issued in core\al\user\ual_cq.c

> If we need to support userspace applications
> posting WRs without a userspace library, then I would much rather take
> the performance hit converting the wr structure when dropping into the
> kernel, versus gunking up the user interface.

Agreed.

>> Your changes in revision 1440 reverted some of the handle
>> exchange\validation between UAL and KAL that shouldn't have been.
>
>  Given what I mentioned above, do you see any actual issues with the
> code?

I don't have issue with the code as part of a larger change, just not on its own like this.  As it stands you reintroduced a bug in there for someone to trip over.  The purpose of the IOCTL-driven WR and CQ processing is to facilitate new HCA drivers being developed.  A functional kernel driver was intended to be able to fully support user-mode clients, though at a performance loss.

So either keep the existing functionality (and fix the IOCTL handling), or delete it.  Don't break the IOCTL handling but leave the functionality in there.  It probably makes sense to have a discussion about removing the WR and CQ IOCTL support before yanking it, though I don't foresee any issues.

-Fab